Risk factors, protective factors, peripheral biomarkers, and neurocognitive markers associated with mood disorders: An umbrella review of 103 meta-analyses and systematic reviews
School of Mental Health and Psychological Science, Anhui Medical University, Hefei, Anhui, China
Department of Psychiatry, The First Affiliated Hospital of Anhui Medical University, Hefei, 230022, China
Hefei Comprehensive National Science Center, Institute of Artificial Intelligence, Hefei, China
Anhui Province Key Laboratory of Cognition and Neuropsychiatric Disorders, Hefei, China
Collaborative Innovation Center of Neuropsychiatric Disorders and Mental Health, Hefei, China
Department of Psychiatry, Chaohu Hospital of Anhui Medical University, Hefei, Anhui, China
Department of Medical Psychology, School of Mental Health and Psychological Science, Anhui Medical University, Hefei, Anhui, China
University of Chinese Academy of Sciences, Beijing, China
⁎Corresponding author. Department of Medical Psychology, School of Mental Health and Psychological Science, Anhui Medical University, Hefei, China. wangkai1964@126.com⁎⁎Corresponding author. Department of Psychiatry, Chaohu Hospital of Anhui Medical University, Hefei, Anhui, China. psyxiaoming@126.comAbstract
Background
Mood disorders, including bipolar disorder (BD) and major depressive disorder (MDD), are extensively studied regarding environmental risk factors, protective factors, peripheral biomarkers, and cognitive neurocognitive markers. Our umbrella review of observational studies aims to separately elucidate these associations with BD and MDD.
Methods
A comprehensive search of PubMed, Embase, Web of Science, and Cochrane was conducted, with a search end date of October 17, 2024. We included systematic reviews that provided meta-analyses of observational studies that examined associations of potential environmental risk factors, environmental protective factors, peripheral biomarkers and neurocognitive markers with mood disorder. We synthesized data using summary effect estimates (e.g., odds ratio [OR], relative risk [RR]), evaluated quality with AMSTAR 2. This review is registered with PROSPERO (CRD42023439581).
Results
Identified 103 articles from 2782 initially screened. In depression, we identified 95 risk factors, 18 protective factors, 19 biomarkers, and 10 neurocognitive markers. For MDD, evidence of association was convincing (class I) history of mental illness (OR 6.77, 95 %CI 5.07–9.04), perinatal domestic violence exposure (OR 3.14, 95 %CI 2.74–3.61), and Mediterranean diet adherence (OR 0.87, 95 %CI 0.84–0.91). For BD, evidence of association was suggestive (class Ⅲ) for cannabis use (OR 2.97, 95 %CI 1.80–4.90), obesity (OR 1.77, 95 %CI 1.40–2.23) were prominent risk factors. The evidence levels of biomarkers and cognitive neuro-markers for BD and MDD are at suggestive (class Ⅲ) or below.
Conclusion
This review highlights the complexity of the interrelationship between mood disorders and numerous risk and protective factors. However, the reviewed associations are not necessarily causal. Future research should prioritize prospective high-quality studies to establish causality and thus facilitate the development of targeted interventions and preventive measures for mood disorders.
Highlights
- •Mediterranean diet and green tea consumption significantly reduce depression risk with Class I evidence.
- •History of mental illness, perinatal domestic violence, job strain, and female gender show Class I evidence.
- •Reduced BDNF levels and MTHFR C677T polymorphism are strongly linked to depression.
- •Marijuana use, rheumatoid arthritis and obesity show suggestive associations.
- •Limited evidence for bipolar protective factors and causality remains unproven for most associations.
1Introduction
Mood disorders, which can be categorized primarily into bipolar and related disorders, with bipolar disorder (BD) being the most prevalent, and depressive disorders, the most common of which is major depressive disorder (MDD) (Cohen and Wills, 1985; Spijker and Claes, 2014), significantly impact mental health. BD is characterized by extreme mood swings, alternating between depressive episodes and manic or hypomanic episodes. The clinical subtypes mainly include BD-I type (with significant manic episodes) and BD-II type (with hypomania and severe depression) (McIntyre et al., 2020). Global epidemiological surveys show that the lifetime prevalence of BD ranges from 1 % to 4 %, and the mortality rate of patients is 2–3 times higher than that of the general population, significantly disrupt daily life, employment, education, physical health, and mental health, potentially leading to severe outcomes such as suicide (Merikangas et al., 2011).
BD is a brain dysfunction disorder caused by the combined effects of genetic susceptibility and environmental factors. Its core mechanism involves imbalances in neurotransmitter systems, energy metabolism abnormalities due to mitochondrial dysfunction, and dysfunction of the prefrontal-limbic neural circuits. The contribution of genetic factors is as high as 80 %. Risk genes such as ANK3 affect the excitability of neurons and calcium signal transduction. At the same time, the neuroinflammatory response induced by chronic stress and the dysfunction of the HPA axis will further disrupt the chemical balance in the brain, leading to the alternating occurrence of depressive and manic states. This multi-system imbalance ultimately manifests as a comprehensive disorder in emotional regulation, cognitive function, and biological rhythms (McIntyre et al., 2020; Merikangas et al., 2011).
MDD is a complex mental disorder caused by the combined effects of genetic susceptibility and environmental stress. Genetic factors interact with environmental stressors (such as chronic stress) by influencing neurotransmitter systems and neural plasticity (such as the decreased expression of BDNF), leading to hyperfunction of the hypothalamic-pituitary-adrenal axis and persistently elevated levels of glucocorticoids. This disorder further triggers neuroinflammatory responses and mitochondrial dysfunction, ultimately causing structural changes and abnormal functional connections in key brain regions, manifesting as core symptoms such as emotional regulation disorders, cognitive impairment, and loss of pleasure (Kendler et al., 2006). MDD, which affects approximately 35 million people globally, is characterized by depressive episodes with no manic or hypomanic episodes (Kendler et al., 2006; Stetler and Miller, 2011).
Notwithstanding extensive research into mood disorders, inconsistencies persist in the literature, especially with regard to markers, risk factors, and protective factors. Although genetic predispositions and environmental influences (e.g., childhood adversity) have been repeatedly identified as risk factors, research findings regarding these factors remain divergent (Taylor, 2017). Research findings on the effectiveness of protective factors such as social support are conflicting, and there are similar disparities in research studies on identifying reliable biological markers such as cortisol levels (Stice et al., 2004). These inconsistencies may arise from variations in research methodologies and the demographics of the study samples (Klok et al., 2011).
Numerous meta-analyses have explored the genetic aspects and protective factors regarding mood disorders, yet the research conclusions remain inconsistent, highlighting the complex interplay between genetic and environmental factors (Moffitt et al., 2005; Sullivan et al., 2000). Similarly, the search for consistent biomarkers continues, with contradictory findings on markers such as cortisol, inflammatory cytokines and BDNF. For instance, several meta-analyses on BD have shown (Mondelli et al., 2010; Risch et al., 2009) that cortisol levels increase during the depressive phase and decrease during the manic phase. This state-dependent change limits its clinical application value as a marker across disease stages. There are also significant differences in the research on BDNF, meta-analysis data showed that serum BDNF levels were mainly decreased in the depressive phase, while some studies in the manic phase reported an increase instead (Risch et al., 2009). These discrepancies underscore the necessity for more comprehensive and integrative research methodologies.
To address these limitations, this umbrella review systematically evaluates the protective factors, environmental risk factors, peripheral biomarkers, and neurocognitive markers associated with mood disorders (Choi and Kang, 2022), collating and assessing epidemiological evidence with a focus on BD and depressive disorders to identify associations with substantial support. This approach facilitates a robust assessment of the strength of evidence substantiating these factors, thus contributing to a clearer understanding of mood disorders (Kim et al., 2019; Theodoratou et al., 2014; Radua et al., 2018).
2Methods
We have followed the Preferred Reporting Items for Overview of Reviews (PRIOR) guidelines on reporting, screening, data extraction, and research methods (e Appendix 1). (Moher et al., 2009; Page et al., 2021) The evaluation of studies for inclusion in this umbrella review was conducted by at least two independent researchers.
2.1Search strategy
We systematically searched PubMed, Embase, Web of Science, and Cochrane databases, covering all available studies in the literature, with a search end date of October 17, 2024. Detailed insights into the search yield are available (e Appendix 2). Two researchers then independently screened titles, abstracts, and full text based on predefined criteria. Any disagreement was solved by consultation between three authors.
2.2Inclusion criteria
This umbrella review includes systematic reviews and meta-analyses. The focus of this review is to explore the protective factors, risk factors, biomarkers and neurocognitive markers related to BD and MDD in mood disorders, as they have the most complete evidence system that could meet the requirements of multi-level evidence in umbrella reviews. Other mood disorders were not included due to insufficient research data. These factors and markers have been defined (e Appendix 3). In instances of multiple papers in the search yield that address the same factor or marker, preference was given to the paper covering the highest number of studies. In cases where the number of studies covered was equal, the paper deemed highest in quality—as determined via a rigorous quality assessment—was selected.
2.3Exclusion criteria
Studies were excluded if they did not investigate potential risk factors, protective factors, biomarkers, or neurocognitive markers of mood disorders. In addition, any study that was not a systematic review or meta-analysis was omitted from the umbrella review. Research involving animal subjects, papers published in languages other than English, and correspondence or summaries drawn from meetings were also excluded.
2.4Data extraction
Two researchers independently extracted the name of the first author, year of publication, sample size, number of cases, study design, effect size, confidence intervals, reports of heterogeneity, and small study effects from the papers that met the inclusion criteria.
2.5Quality assessment
We utilized AMSTAR 2 (Shea et al., 2017), a comprehensive and reliable tool for assessing the quality of systematic reviews and meta-analyses (e Appendix 5). Utilizing the evaluation criteria in the AMSTAR 2, the quality of the papers considered for inclusion in the umbrella review was rigorously evaluated (e Appendix 6). It is important to note that the AMSTAR 2 assessment results are not used for screening or excluding any studies. All studies that meet the inclusion criteria have been included. This assessment was conducted with the aim of: (1) comprehensively describing the overall methodological rigor of the included systematic reviews. (2) exploring important background information on the potential impact of the quality of the included studies on the overall results. (3) identifying the methodological advantages and limitations in the existing evidence.
2.6Grading the strength of evidence
The papers included in this umbrella review were classified into five distinct categories based on several criteria (e Appendix 7). The criteria included heterogeneity, the number of cases, the p-value under a random effects model, 95 % prediction intervals, the presence of any small study effects, and evidence of excessive significance bias (Kim et al., 2019; Theodoratou et al., 2014; Radua et al., 2018).
2.7Data analysis
We recalculated the combined effect sizes and confidence intervals under a random effects model. In addition, we computed 95 % prediction intervals for the effect sizes, yielding an estimate of the potential range of these sizes in future studies. The presence of small study effects bias was investigated when the Egger's p-value was less than 0.1. Excessive significance bias was assessed by comparing the actual number of statistically significant findings to the expected number. In line with evidence rating standards, we classified the relevant factors or markers into one of the five classes determined in our data analysis (Table 1). For the data analysis, Comprehensive Meta-Analysis 3.0, Python 3.1, and GraphPad Prism version 8.0 were employed as software tools (Amani and Amani, 2023).Random effects p-value Number of cases p value of the largest study Heterogeneity(I2) Small study effects Excessive significance bias 95 % prediction interval Convincing (class I) >1000 <0.000001 Not detected <50 % No No Excluding null Highly suggestive (class Ⅱ) >1000 <0.000001 … … … … … Suggestive (class Ⅲ) >1000 <0.001 … … … … … Weak (class Ⅳ) … >0.05 … … … … … Not significant (class NS) … >0.05 … … … … …
3Results
3.1Search yield
By October 17, 2024, from an initial pool of 2914 papers, we had determined that 103 papers met our umbrella review inclusion criteria, and these papers were subsequently analyzed (Fig. 1).
3.2Risk factors for major depressive disorder
Among the 103 papers included in this umbrella review, 61 studies identify a total of 95 environmental risk factors for depression (Table 2). These 61 studies collectively cover 8837 cases, with a median of 1638 depression cases per meta-analysis, and a range of 1014 to 4548 cases. The studies analyzed in this review have a total sample size of 74,107, with a median of 8,406, and a range of 612 to 4268. Based on our criteria for evidence strength, 4 of these factors were classified as level 1 (Table 3, Fig. 2, e Appendix 8), indicating the highest degree of evidence for their association with depression. These factors, p-value under a random effects model being 0.000001. Furthermore, the heterogeneity (I2) for these studies was less than 50 %. The prediction intervals and 95 % confidence intervals for each factor consistently excluded null values, indicating the absence of small study effects. The Egger's value for each factor exceeded 0.1, indicating the absence of an overt significance bias.Potential environmental risk factors of depression Potential environmental protective factors of depression Potential peripheral biomarkers of depression Potential neurocognitive makers of depression Potential environmental risk factors of bipolar disorder Potential peripheral biomarkers of bipolar disorder Potential neurocognitive makers of bipolar disorder Number of articles 61 11 11 5 4 7 4 Number of factors or markers 95 18 19 10 6 15 5 Cases 8837 3634208 14395 9675 124365 2680 20745 Median cases count 1638 1462 356 608 3693 99 2197 Range of cases 1014 to 4538 286 to 3634208 75 to 6098 361 to 2453 586 to 91593 52 to 548 251 to 15752 Population 74107 5125472 82106 22202 3687969 13458 64085 Median population count 8406 23107 9433 1594 47464 545 3055 Range of population 612 to 4268 1404 to 3920383 467 to 45110 970 to 5600 2375 to 2957895 104 to 4598 396 to 50563 Study Number of case/total population Number of studies Study design Effect size Estimate (95 %CI) Random effects p value I2 95 % prediction interval Egger's P value Large heterogeneity,
small study effect,
excess significance
biasAMSTAR 2 quality Convincing (class I) History of mental illness Ansari et al. (2021) 1543/3515 8 Cross-sectional,
cohortOR 6.77(5.07,9.04) p < 0.000001 0 % 6.48 to 7.05 0.941 None Critically low Perinatal domestic
violenceHoward et al. (2013) 1732/14914 3 Cohort RR 3.14(2.74,3.61) p < 0.000001 0 % 2.96 to 3.23 0.18 None Low Job strain Madsen et al. (2017) 1014/27461 7 Cohort RR 1.77(1.47,2.13) p < 0.000001 24 % 1.58 to 1.95 0.66 None Critically low Female Tang et al. (2014) 4548/28217 20 Cross-sectional,
case-control,
cohortOR 1.57(1.39, 1.79) p < 0.000001 0 % 1.44 to 1.69 0.73 None Critically low Highly suggestive (class Ⅱ) Poor marital relationship Qi et al. (2021) 2582/10348 17 Case-control,
cohortOR 3.56(2.95,4.28) p < 0.000001 30 % 3.37 to 3.74 0.067 Large heterogeneity;
small study effectCritically low History of common male diseases Dadi et al. (2020) 6789/11799 13 Cross-sectional,
case-control,
cohortOR 3.27(2.47,4.33) p < 0.000001 89 % 2.98 to 3.55 0.34 Large heterogeneity Critically low Premature infants Eduardo et al. (2019) 8231/406485 12 Cross-sectional,
case-control,
cohortOR 3.16(2.18,4.58) p < 0.000001 54 % 2.78 to 3.53 0.85 Small study effect Critically low Parents passed away Geulayov et al. (2012) 700394/2097859 20 Cohort RR 3.01(2.81,3.23) p < 0.000001 52 % 2.94 to 3.08 NR Large heterogeneity Critically low History of Violence Dadi et al. (2020) >1000/7428 11 Cross-sectional,
case-control,
cohortOR 2.99(2.20,4.07) p < 0.000001 71 % 2.68 to 3.29 0.886 Large heterogeneity Critically low Between mother-in-law
and daughter-in-lawQi et al. (2021) 1400/3582 9 Case-control, cohort OR 2.89(2.12,3.95) p < 0.000001 72 % 2.58 to 3.09 0.246 Large heterogeneity;
excess significance biasCritically low Sexual abuse Chen et al. (2010) 1099/4897 16 Case-control OR 2.66(2.14,3.30) p < 0.000001 57 % 2.44 to 2.88 0.92 Large heterogeneity Low Lack of social support Qi et al. (2021) 1696/9728 9 Case-control,
cohortOR 2.57(2.32,2.85) p < 0.000001 93 % 2.46 to 2.67 0.578 Large heterogeneity;
excess significance biasCritically low Child abuse Gardner et al. (2019) 61867/124960 21 Cross-sectional,
case-control,
longitudinalOR 2.48(2.14,2.87) p < 0.000001 51 % 2.33 to 2.63 0.001 Large heterogeneity,
small study effectLow Insomnia Li et al. (2016) 85538/172077 34 Cohort RR 2.27(1.89,2.71) p < 0.000001 92 % 2.08 to 2.45 0.013 Large heterogeneity;
small study effect;
excess significance biasCritically low Premenstrual syndrome Cao et al. (2019) 4385/8990 19 Cross-sectional,
case-control,
cohortOR 2.20(1.81,2.68) p < 0.000001 42 % 2.00 to 2.39 0.0015 Small study effect;
excess significance biasLow Epilepsy Chu et al. (2022) 2752/15699 23 Cross-sectional,
cohortOR 2.05(1.77,2.37) p < 0.000001 37 % 1.86 to 2.23 0.001 Small study effect Critically low Violence Zhang et al. (2019) 74958/177531 32 Cohort,
longitudinalOR 2.04(1.72,2.41) p < 0.000001 93 % 1.86 to 2.21 0.2 Large heterogeneity Critically low Financial difficulty Dadi et al. (2020) 4560/11207 14 Cross-sectional,
case-control,
cohortOR 2.03(1.63,2.53) p < 0.000001 74 % 1.81 to 2.24 0.31 Large heterogeneity Critically low Intimate partner violence Devries et al. (2013) 63170/124960 6 Longitudinal OR 1.97(1.56,2.48) p < 0.000001 50 % 1.74 to 2.20 0.12 Large heterogeneity Critically low Chronic diseases in the elderly Huang et al. (2009) 1510/13864 12 Longitudinal RR 1.84(1.49,2.26) p < 0.000001 73 % 1.63 to 2.05 0.98 Large heterogeneity Critically low Infection with COVID Lin et al. (2013) 3103/13800 13 Case-control,
cohortOR 1.83(1.70,1.97) p < 0.000001 96 % 1.75 to 1.90 0.432 Large heterogeneity Critically low Urinary incontinence Cheng et al. (2020) 6244/31462 12 Cross-sectional,
cohortOR 1.73(1.64,1.82) p < 0.000001 75 % 1.68 to 1.78 0.05 Large heterogeneity,
small study effectLow Eczema Long et al. (2022) 141910/4736222 18 Case-control,
cohortOR 1.64(1.39,1.94) p < 0.000001 90 % 1.47 to 1.80 0.233 Large heterogeneity Critically low Elderly people have low levels of education Huang et al. (2009) 5857/24067 24 Cross-sectional OR 1.58(1.38,1.82) p < 0.000001 74 % 1.44 to 1.72 0.92 Large heterogeneity Critically low Unmarried YAN et al. (2011) 4885/24044 24 Cross-sectional OR 1.55(1.37,1.74) p < 0.000001 74 % 1.42 to 1.67 0.0007 Large heterogeneity;
small study effectCritically low Cardiovascular risk factors Valkanova (2013) 8076/17899 18 Cross-sectional,
longitudinalOR 1.49(1.27,1.75) p < 0.000001 76 % 1.33 to 1.64 0.06 Large heterogeneity;
small study effectCritically low Night shift work Lee et al. (2017) 47138/94277 11 Cross-sectional,
longitudinal,
cohortOR 1.43(1.24,1.64) p < 0.000001 78 % 1.28 to 1.57 0.088 Large heterogeneity,
small study effectCritically low Arthritis Xue et al. (2020) 145865/629077 14 Case-control HR 1.42(1.34,1.52) p < 0.000001 0 % 1.36 to 1.47 NA Excess significance bias Critically low Neuroticism Puyan'e et al. (2022) 5054/10908 13 Cross-sectional,
case-control,
cohortOR 1.37(1.22,1.53) p < 0.000001 88 % 1.48 to 1.61 0.0006 Large heterogeneity;
small study effectLow Adverse obstetric history Dadi et al. (2020) 6789/13450 16 Cross-sectional,
case-control,
cohortOR 2.01(1.67,2.42) p < 0.000001 81 % 1.80 to 2.21 0.11 Large heterogeneity Low Second-hand smoke Han et al. (2019) 600256/1290873 24 Cohort,
cross-sectionalOR 1.32(1.25,1.39) p < 0.000001 72 % 1.26 to 1.37 0.19 Small study effect Low Suggestive (class Ⅲ) Marijuana Gibbs et al. (2015) 5520/9373 2 Cohort RR 2.97(1.80,4.90) p < 0.001 0 % 2.46 to 3.47 0.14 None Low Have Complications of diabetes Simayi et al. (2019) 29264419/79307024 14 Cross-sectional,
case-control,
cohortOR 2.91(1.76,4.82) p < 0.001 49 % 2.40 to 3.41 0.11 None Critically low Elderly frailty Soysal et al. (2017) 1278/8023 5 Cross-sectional OR 2.64(1.59,4.37) p < 0.001 55 % 2.13 to 3.14 0.98 Large heterogeneity Critically low Financial instability Ansari et al. (2021) 1686/3052 5 Cross-sectional,
cohortOR 2.24(1.44,3.48) p < 0.001 74 % 1.79 to 2.68 0.0096 Large heterogeneity;
small study effectCritically low Diabetes Qiu et al. (2020) 1598/10067 4 Case-control OR 2.00(1.36,2.64) p < 0.001 0 % 1.61 to 2.38 0.87 None Critically low Not receiving secondary education Simayi et al. (2019) 1134/4767 8 Cross-sectional,
case-control,
cohortOR 1.91(1.30,2.81) p < 0.001 84 % 1.52 to 2.29 0.98 None Low Periodontal disease Cademartori et al. (2018) 12708/63540 6 Longitudinal HR 1.73(1.58,1.89) p < 0.001 0 % 1.33 to 1.64 NA None Low Unemployment during pregnancy Luo et al. (2022) 2111/11256 9 Case-control OR 1.68(1.25,2.25) p < 0.001 56 % 1.39 to 1.96 0.171 Large heterogeneity Low Mother's depression Ansari et al. (2021) 3654/6661 7 Cross-sectional,
cohortOR 1.66(1.27,2.17) p < 0.001 88 % 1.39 to 1.92 0.016 Large heterogeneity;
small study effectCritically low Gestational diabetes Azami et al. (2019) 156793/2370958 18 Cross-sectional,
case-control,
cohortRR 1.59(1.22,2.07) p < 0.001 0 % 1.32 to 1.85 0.05 Small study effect Moderate Exposure to effort reward imbalance at work Rugulies et al. (2017) 2897/84963 8 Cohort RR 1.49(1.23,1.80) p < 0.001 59 % 1.29 to 1.68 0.79 Large heterogeneity Low Low level of education Yang et al. (2022) 8524/32479 24 Case-control OR 1.40(1.18,1.67) p < 0.001 85 % 1.22 to 1.57 0.057 Large heterogeneity;
small study effect;
excess significance biasCritically low Insufficient support from relatives Dadi et al. (2020) 5687/11654 9 Case-control,
cohortOR 1.36(1.16,1.56) p < 0.001 78 % 1.20 to 1.51 0.00003 Large heterogeneity;
small study effect;
excess significance biasCritically low Metabolic syndrome Kim et al. (2023) NA 11 Cohort RR 1.29(1.12,1.48) p < 0.001 79 % 1.14 to 1.43 NA Large heterogeneity Critically low Inflammatory bowel disease Barberio et al. (2021) 16911/29438 26 Cross-sectional,
cohortOR 1.20(1.10,1.40) p < 0.001 23 % 1.11 to 1.29 0.28 None Critically low Red meat and processed meat intake Nucci et al. (2020) 192185/241738 17 Cross-sectional,
case-controlOR 1.08(1.04,1.12) p < 0.001 77 % 1.04 to 1.11 0.777 Large heterogeneity High Weak (class Ⅳ) Poor self-perceived health status Qiu et al. (2020) 789/4919 3 Case-control OR 7.99(1.07,14.9) p < 0.05 78 % 5.97 to 10.00 NA Large heterogeneity Critically low Antenatal Depression Qi et al. (2021) 524/3303 9 Case-control
cohortOR 7.70(6.02,9.83) p < 0.000001 69 % 7.45 to 7.94 0.488 Large heterogeneity;
excess significance biasCritically low Marital conflict Zegeye et al. (2018) 690/1358 4 Case-control,
cohortOR 7.20(2.73,19.02) p < 0.001 83 % 6.21 to 8.18 0.351 Large heterogeneity;
excess significance biasCritically low Father unemployed Ansari et al. (2021) 758/1505 3 Cross-sectional,
cohortOR 6.61(1.94,22.54) p < 0.05 59 % 5.38 to 7.83 0.359 Large heterogeneity Critically low Unfavorable marital status Dadi et al. (2020) 6802/12935 6 Case-control,
cohortOR 4.17(1.75,9.94) p < 0.05 81 % 3.30 to 5.03 0.0012 Large heterogeneity;
small study effectLow Poor economic foundation Qi et al. (2021) 725/5393 12 Case-control
cohortOR 3.67(3.07,4.37) p < 0.000001 22 % 3.49 to 3.84 0.657 Excess significance bias Low Low level of education Ansari et al. (2021) 897/1697 2 Cross-sectional,
cohortOR 3.56(1.06,11.97) p < 0.05 87 % 2.34 to 4.77 0.639 Large heterogeneity Critically low History of Complications of pregnancy Zegeye et al. (2018) 173/1324 3 Case-control,
cohortOR 3.21(1.77,5.81) p < 0.001 65 % 2.62 to 3.77 0.573 Large heterogeneity Critically low Parental separation SIMBI et al. (2020) 268/758 7 Case-control OR 3.14(1.92,5.15) p < 0.001 0 % 2.64 to 3.63 0.395 None Critically low Have a history of miscarriage Zegeye et al. (2018) 765/1413 3 Case-control,
cohortOR 3.03(2.11,4.37) p < 0.000001 92 % 2.64 to 3.35 0.568 Large heterogeneity;
excess significance biasCritically low Male Gender Preference Dadi et al. (2020) 657/1135 4 Cross-sectional,
case-control,
cohortOR 2.97(1.41,6.26) p < 0.05 88 % 2.25 to 3.71 NA Large heterogeneity Low Low concentration of vitamin D Aghajafari et al. (2018) 796/2879 5 NR OR 2.72(1.42,5.22) p < 0.05 0 % 2.71 to 3.36 NA None Critically low Unplanned pregnancy Qi et al. (2021) 619/3609 7 Case-control
cohortOR 2.55(2.08,1.14) p < 0.000001 26 % 2.34 to 2.75 0.139 Large heterogeneity;
excess significance biasLow Poor living condition Qi et al. (2021) 784/3021 6 Case-control
cohortOR 2.44(1.92,3.10) p < 0.000001 26 % 2.20 to 2.67 0.013 Large heterogeneity;
small study effectLow Negative life events Qiu et al. (2020) 598/10067 4 Case-control OR 2.43(1.87,2.98) p < 0.000001 65 % 2.16 to 2.69 NA Large heterogeneity Critically low Previous trauma Tang et al. (2014) 8900/17890 16 Cross-sectional,
case-control,
cohortOR 2.26(1.34,3.81) p < 0.05 67 % 1.73 to 2.78 NA Large heterogeneity Critically low Parental loss SIMBI et al. (2020) 716/2068 9 Case-control OR 2.18(1.63,2.90) p < 0.000001 15 % 1.88 to 2.47 0.895 None Low Suffering from chronic physical diseases before pregnancy Luo et al. (2022) 1670/7678 5 Case-control OR 2.10(1.13,3.90) p < 0.05 89 % 1.48 to 2.71 0.383 Large heterogeneity High My mother-in-law is a caregiver Qi et al. (2021) 527/2258 5 Case-control
cohortOR 1.95(1.54,2.48) p < 0.000001 33 % 1.71 to 2.18 0.271 Large heterogeneity;
excess significance biasCritically low Not graduating from high school Simayi et al. (2019) 697/1767 8 Cross-sectional,
case-control,
cohortOR 1.91(1.30,2.81) p < 0.001 84 % 1.52 to 2.29 0.98 Large heterogeneity Low Myocytopenia Chang et al. (2017) 1788/20753 6 Cohort,
cross-sectionalOR 1.82(1.16,2.59) p < 0.05 74 % 1.37 to 2.27 0.019 Small study effect Critically low Dementia patients Huang et al. (2011) 10700/21400 17 Cross-sectional,
cohortOR 1.82(1.15,2.89) p < 0.05 55 % 1.36 to 2.27 0.38 Large heterogeneity Critically low Postpartum stress events Ansari et al. (2021) 678/1731 2 Cross-sectional,
cohortOR 1.79(1.37,2.35) p < 0.001 95 % 1.77 to 1.79 0.286 Large heterogeneity Low Parental death SIMBI et al. (2020) 318/1058 5 Case-control OR 1.76(1.13,2.73) p < 0.05 0 % 1.37 to 2.18 0.532 None Low Premature ejaculation Xia et al. (2016) >1000/18035 8 Cross-sectional,
cohortOR 1.63(1.42,1.87) p < 0.000001 0 % 1.49 to 1.76 0.017 Small study effect Low Using axle neck for labor analgesia for parturients in low incidence rate areas Wang et al. (2022) 1876/4439 7 Cohort RR 1.56(1.16,2.10) p < 0.05 23 % 1.26 to 1.85 0.964 None Critically low Sleep disorder in children and adolescents Marino et al. (2021) 6503/13006 8 Cohort RR 1.50(1.13,2.00) p < 0.05 50 % 1.21 to 1.78 0.002 Large heterogeneity Critically low Insulin use in patients with diabetes Bai et al. (2018) 331085/4912160 12 Cohort,
cross-sectionalOR 1.41(1.13,1.76) p < 0.05 69 % 1.19 to 1.63 0.94 Large heterogeneity Low Undereducated Luo et al. (2022) 1896/8606 6 Case-control OR 1.41(1.10,1.81) p < 0.05 54 % 1.16 to 1.65 0.575 Large heterogeneity High Hypothyroidism Bode et al. (2021) 14578/348 014 25 Cross-sectional,
cohortOR 1.30(1.08,1.57) p < 0.05 74 % 1.11 to 1.48 NA Large heterogeneity High Edentulous disease Cademartori et al. (2018) 91417/182835 5 Cross-sectional OR 1.28(1.06,1.55) p < 0.05 0 % 1.26 to 1.47 0.47 None Low Birth weight less than 2500 g Shintani et al. (2023) 1865/85555 5 Case-control OR 1.28(1.04,1.56) p < 0.05 0 % 1.07 to 1.48 0.18 None High Parenting self-efficacy Ansari et al. (2021) 1987/3424 2 Case-control,
cohortOR 1.27(0.73,2.20) p < 0.001 71 % 0.71 to 1.82 0.258 Large heterogeneity Critically low Type 2 diabetes Nouwen et al. (2010) >1000/48808 11 Cross-sectional,
case-control,RR 1.24(1.09,1.40) p < 0.05 67 % 1.11 to 1.37 0.81 Large heterogeneity Critically low Unhealthy obesity Jokela et al. (2013) 15168/30337 8 Cohort RR 1.23(1.05,1.45) p < 0.05 4 % 1.07 to 1.38 NA None Critically low Unhealthy eating patterns Li et al. (2017) 7331/20363 19 Cross-sectional,
case-control,
cohortOR 1.18(1.05,1.34) p < 0.05 71 % 1.05 to 1.30 0.073 Large heterogeneity;
small study effectCritically low Sedentary behavior during screen time Liu et al. (2015) 63857/127714 16 Cross-sectional,
cohortOR 1.12(1.03,1.22) p < 0.05 82 % 1.03 to 1.20 0.642 Large heterogeneity High Air pollution Braithwaite et al. (2019) 5545/84619 5 Cross-sectional;
cohortOR 1.10(1.02,1.19) p < 0.05 0 % 1.02 to 1.18 0.311 None High Reduced general sense of support Luo et al. (2022) 627/4489 3 Case-control OR 1.06(1.03,1.10) p < 0.001 0 % 1.03 to 1.08 NA None Low Parenting stress Ansari et al. (2021) 356/692 2 Cross-sectional,
cohortOR 1.06(1.02,1.11) p < 0.05 5 % 1.02 to 1.09 0.383 None Critically low Homozygous mutation of rs41423247 gene polymorphism of Glucocorticoid receptor Peng et al. (2018) 1630/4992 9 Case-control OR 0.77(0.64,0.94) p < 0.05 32 % 0.58 to 0.95 0.31 None Critically low Perinatal smoking Yang et al. (2022) 123/851 6 Case-control OR 0.63(0.45,0.87) p < 0.05 26 % 0.29 to 0.96 0.0045 Large heterogeneity;
small study effect;
excess significance biasCritically low Drink Yang et al. (2022) 63/370 5 Case-control OR 0.43(0.23,0.80) p < 0.05 66 % 0.19 to 1.05 NA Large heterogeneity Critically low Not significant (NS) Family environment Pedersen et al. (2022) 3809/5980 3 Cohort RR 2.30(0.73,7.24) p > 0.05 66 % 1.15 to 3.44 0.722 Large heterogeneity Critically low Parent depression Pedersen et al. (2022) 6509/18730 8 Cohort RR 1.60(0.82,3.10) p > 0.05 89 % 0.93 to 2.26 0.0034 Large heterogeneity;
small study effectCritically low Hypertension Long et al. (2014) 568/9647 5 Cohort RR 1.16(0.91,1.42) p > 0.05 65 % 0.91 to 1.40 0.221 Large heterogeneity low Toxoplasma infection Chegeni et al. (2019) 2968/25580 29 Cross-sectional,
case-controlOR 1.15(0.95,1.39) p > 0.05 80 % 0.96 to 1.34 0.148 Large heterogeneity Critically low Spouses of long-term cancer survivors and healthy group spouses Mitchell et al. (2013) 51381/217630 16 Cross-sectional,
case-control,
cohortRR 1.11(0.96,1.27) p > 0.05 84 % 0.96 to 1.26 0.88 Large heterogeneity Critically low
3.3Protective factors for major depressive disorder
Eleven papers that met the criteria for examining 18 potential environmental protective factors against depression (Table 2) were included in this umbrella review. The total case count was 3,634,208, with individual case counts, with a range of 286 to 3,634,208 and a median of 1462. Our analysis covered a total sample size of 5,125,472, with a median of 23,107 and a range of 1404 to 3,920,383. Based on our criteria for evidence strength, 2 of the 18 potential environmental protective factors were classified as having level 1 associations with depression (Table 4, Fig. 3, e Appendix 9). Regarding statistical significance, 4 factors (22.22 %) had a random effect p-value of 0.000001, 1 factor (1.56 %) had a random effect p-value of 0.001, and 12 factors (66.67 %) had a random effect p-value of 0.05. With respect to heterogeneity, 6 (33.33 %) of the factors exhibited an I2 of 50 %. Notably, the prediction intervals and 95 % confidence intervals of 17 (94.44 %) of the factors consistently excluded null values, demonstrating robustness. The Egger's values exceeded 0.1 for 11 (61.11 %) of the factors, indicating the absence of small study effects and excessive significance bias across the studies.Study Number of case/ total population Number of studies Study design Effect size Estimate (95 %CI) Random effects p value I2 95 % prediction
intervalEgger's
P valueLarge heterogeneity,
small study effect,
excess significance
biasAMSTAR2 quality/
AMSTAR2 quality
when protocol
assessment was
ruled outConvincing (class I) Mediterranean diet score (MDS) Nicolaou et al. (2019) 10563/23107 6 Cohort,
cross-sectionalOR 0.87(0.84,0.91) p < 0.000001 0 % 0.83 to 0.90 0.37 None Critically low Green tea consumption Yaegashi et al. (2022) 1192/11820 8 Cross-sectional,
cohortOR 0.66 (0.58,0.74) p < 0.000001 0 % 0.53 to 0.78 0.815 None Critically low Highly suggestive (class Ⅱ) Fruit consumption Liu et al. (2016) >1000/227852 10 Cohort,
cross-sectionalOR 0.86(0.81,0.91) p < 0.000001 48 % 0.80 to 0.91 0.07 Small study effect Critically low Gender-male Simayi et al. (2019) 3567976/3920383 10 Case-control OR 0.56(0.47,0.65) p < 0.000001 69 % 0.38 to 0.73 0.5597 Large heterogeneity Low Suggestive (class Ⅲ) Dietary fiber intake Saghafian et al. (2021) 1732/14914 18 Cohort,
cross-sectional,
case-controlOR 0.90(0.86,0.95) p < 0.001 67 % 0.85 to 0.94 0.001 Large heterogeneity,
small study effectModerate Weak (class Ⅳ) Alternative Healthy Diet Index Nicolaou et al. (2019) 8079/23107 6 Cohort,
cross-sectionalOR 0.93(0.88,0.98) p < 0.05 34 % 0.87 to 0.98 0.079 Small study effect Critically low Good social support Qiu et al. (2020) >1000/9106 4 Cross-sectional OR 0.91(0.84,0.97) p < 0.05 61 % 0.82 to 0.99 NA Large heterogeneity Critically low Vegetable consumption Liu et al. (2016) >1000/218699 8 Cohort,
cross-sectionalOR 0.89(0.83,0.94) p < 0.05 14 % 0.83 to 0.94 0.2 None Critically low Fermented milk products intake Luo et al. (2023) 6118/76146 8 Cohort RR 0.89(0.81,0.98) p < 0.05 67 % 0.79 to 0.98 0.508 Large heterogeneity Low Fish consumption Li et al. (2015) 10961/150278 26 Cohort,
cross-sectionalRR 0.83(0.74,0.93) p < 0.05 64 % 0.71 to 0.94 0.419 Large heterogeneity Critically low Higher levels of vitamin D Aghajafari et al. (2018) 796/NR 5 Case-control OR 0.81(0.70,0.92) p < 0.05 0 % 0.66 to 0.95 NA None Critically low Coffee Grosso et al. (2016) 5253/327697 7 Cross-sectional,
cohortRR 0.76(0.64,0.91) p < 0.05 71 % 0.55 to 0.93 0.00257 Large heterogeneity,
small study effectLow Vegetable intake Saghafian et al. (2018) 6543/90182 13 Cohort,
cross-sectional,
case-controlRR 0.76(0.63,0.92) p < 0.05 83 % 0.57 to 0.94 0.055 Large heterogeneity,
small study effectCritically low Current social situation Simayi et al. (2019) 453/1574 7 Case-control OR 0.64(0.47,0.88) p < 0.05 69 % 0.33 to 0.94 0.554 Large heterogeneity Low Effect of axle neck analgesia on parturients in areas with high incidence rate of PPD Wang et al. (2022) 1167/3446 9 Cohort RR 0.61(0.39,0.94) p < 0.05 79 % 0.16 to 1.05 0.104 Large heterogeneity Critically low Marital status (married) Simayi et al. (2019) 1013/2650 7 Case-control OR 0.53(0.34,0.83) p < 0.05 75 % 0.08 to 0.97 0.6069 Large heterogeneity Low Take regular exercise Simayi et al. (2019) 286/1404 5 Case-control OR 0.51(0.27,0.96) p < 0.05 91 % 0.12 to 1.14 0.1948 Large heterogeneity Critically low Not significant (NS) Dietary methods to stop hypertension Nicolaou et al. (2019) 9076/23107 6 Cohort,
cross-sectionalOR 0.94(0.87,1.01) p > 0.05 81 % 0.86 to 1.01 0.5257 Large heterogeneity Critically low
3.4Biomarkers of major depressive disorder
Our review of biomarkers of depression covers findings from 11 eligible papers on studies that investigate 19 potential biomarkers of depression (Table 2). The total case count for these studies was 14,395, with a median case count of 356, and a range of 75 to 6098. The total sample size across these studies was 82,106, with a median of 943, and a range of 467 to 45,110. On average, each biomarker was studied in nine published studies (range: 2–24). Regarding statistical significance (Table 5, Fig. 4, e Appendix 10), 3 (15.79 %) of the biomarkers had a random effect p-value of 0.05, 7 (36.84 %) had a random effect p-value of 0.01, and 9 (47.37 %) biomarkers were not statistically significant. The heterogeneity among these studies was generally low, with an I2 of more than 50 % reported for only 3 (15.79 %) of the significant biomarkers. The prevalence of small study effects was minimal, as evidenced by 8 (80 %) of the factors being the subjects of studies with an Egger's value greater than 0.1, which indicated robust findings. None of the studies exhibited excessive bias toward significance.Study Number of case/total population Number of studies Study design Effect size Estimate
(95 %CI)Random effects
p valueI2 95 % prediction
intervalEgger's
P valueLarge heterogeneity,
small study effect,
excess significance
biasAMSTAR2 quality/
AMSTAR2 quality
when protocol
assessment was
ruled outSuggestive (class Ⅲ) BDNF Tiwari et al. (2022) 1130/2580 24 Case-control SMD −0.89(-1.41,-0.38) p < 0.001 96 % −1.41 to −0.37 0.002 Large heterogeneity,
small study effectCritically low Weak (class Ⅳ) Plasma lipids Liu et al. (2020) 567/1014 9 Cross-sectional,
case-control,
cohortSMD 1.58(0.75,2.41) p < 0.001 96 % 0.75 to 2.4 0.523 Large heterogeneity Critically low Serum Li et al. (2019) 1157/16287 6 Cohort HR 0.88(0.78,0.99) SMD = −0.070(-0.136, −0.005) 0.006 79 % 0.78 to 0.97 0.94 Large heterogeneity Moderate Triglyceride Bharti et al. (2021) 869/2588 10 Case-control SMD 0.55(0.30,0.80) p < 0.001 62 % 0.30 to 0.79 0.13 Large heterogeneity Critically low VEGF Carvalho et al. (2015) 287/1633 14 Case-control Hedges' g 0.34(0.14,0.54) p < 0.001 66 % 0.14 to 0.53 0.746 Large heterogeneity Critically low 8-Hydroxy-2′-deoxyguanosine Black et al. (2014) 132/579 10 Case-control,
cohortHedges' g 0.31(0.06,0.60) 0.01 75 % 0.06 to 0.55 0.69 Large heterogeneity Low CSF HVA Ogawa et al. (2018) 419/803 23 Cross-sectional Hedges' g −0.37(-0.55,-0.19) 0.000061 25 % −0.54 to −0.19 0.256 None Critically low Very low density lipoprotein Bharti et al. (2021) 170/635 2 Case-control SMD −0.46(-0.72,-0.21) p < 0.05 0 % −0.71 to −0.20 0.25 None Critically low GDNF Lin et al. (2015) 526/1028 13 Cross-sectional Hedges' g −0.62(-0.99,-0.25) p < 0.001 86 % −0.98 to −0.25 0.47 Large heterogeneity Critically low TRP Liu et al. (2022) 140/690 5 Cohort SMD −5.30(-7.72,-3.05) p < 0.001 0 % −7.72 to −3.05 0.0016 None Low Not significant (NS) Docosahexaenoic acid Bharti et al. (2021) 209/578 2 Case-control SMD 2.89(-44.17,49.96) 0.903 98 % −44.13 to 49.9 0.61 Large heterogeneity Low Lipid peroxidation Brown et al. (2014) 75/943 12 Case-control Hedges' g 1.62(1.02,2.22) 0.166 93 % 1.02 to 3.61 0.983 Large heterogeneity Low Glycosylated hemoglobin Beran et al. (2022) 6098/45110 5 Longitudinal OR 1.19(1.08,1.32) SMD = 0.096(0.041, 0.151) p > 0.05 0 % 1.09 to 1.28 0.271 None Low Apolipoprotein A Bharti et al. (2021) 356/876 2 Case-control SMD 0.61(-1.31,2.54) 0.532 25 % −0.13 to 2.52 0.11 None Critically low Eicosapentaenoic acid Bharti et al. (2021) 473/956 3 Case-control SMD 0.11(-1.43,1.66) 0.888 85 % −1.42 to 1.64 0.19 Large heterogeneity Critically low High density lipoprotein Bharti et al. (2021) 290/782 11 Case-control SMD −0.26(-0.70,0.18) 0.246 93 % −0.69 to 0.17 0.36 Large heterogeneity Critically low Total cholesterol Bharti et al. (2021) 1234/4081 16 Case-control SMD −0.46(-0.93,0.01) p > 0.05 93 % −0.92 to 0.01 0.05 Large heterogeneity Critically low Apolipoprotein B Bharti et al. (2021) 154/467 2 Case-control SMD −0.47 (−1.06,0.11) 0.117 42 % −0.10 to 0.11 0.30 None Critically low Omega-3 fatty acid Bharti et al. (2021) 109/476 2 Case-control SMD −2.01(-17.84,13.82) 0.903 98 % −17.82 to 13.80 0.16 Large heterogeneity Low
3.5Neurocognitive markers of major depressive disorder
Our review of neurocognitive markers of depression (Table 2) examines findings from five papers, each of which meets the inclusion criteria and investigates 10 potential neurocognitive markers of depression. The total number of cases across these studies was 9,657, with a median case count of 608 and a range of 361 to 2453. The total sample size was 22,202 individuals, with a median of 1,594, and a range of 970 to 5600. Regarding statistical significance (Table 6, Fig. 5, e Appendix 11), 2 (20 %) of the 10 neuromarkers had a random effect p-value of 0.001, 3 neuromarkers (30 %) had a p-value of 0.05, and 5 (50 %) were not statistically significant (p > 0.05). With respect to heterogeneity, 3 (60 %) of the significant factors had an I2 of more than 50 %, indicating moderate heterogeneity among these specific studies. The Egger's value of each study exceeded 0.1, indicating the absence of publication bias. Neither excessive significance bias nor small study effects were observed in these studies.Study Number of case/total population Number of studies Study design Effect size Estimate
(95 %CI)Random effects
p valueI2 95 % prediction
intervalEgger's
P valueLarge heterogeneity,
small study effect,
excess significance
biasAMSTAR2 quality/
AMSTAR2 quality
when protocol
assessment was
ruled outSuggestive (class Ⅲ) MTHFR C677T polymorphism Jiang et al. (2015) 1895/3808 13 Case-control OR TT vs. CC:
2.19(1.49,3.24);
TT vs. CC + CT:
1.80(1.31,2.46);
TT + CT vs. CC:
1.64(1.16,2.30)
T vs C:
1.52(1.24,1.85);p < 0.001 74 % 1.31 to 1.72 0.1345 Large heterogeneity Critically low Weak (class Ⅳ) 5-HTTLPR S-Allele in the additive model Cheng et al. (2014) 361/970 3 Cohort,
case-control,
cross-sectionalOR 2.64(1.01,6.88) p < 0.05 55 % 1.67 to 3.60 0.475 Large heterogeneity Critically low Frontal lobe lesion Douven et al. (2017) NR 35 NR OR 1.54(91.27,1.88) p < 0.001 43 % 1.34 to 1.73 0.92 None Moderate 5-HTTLpR S-Allele in the recessive model Cheng et al. (2014) 456/970 4 Cohort,
case-control,
cross-sectionalOR 1.43(1.08,1.90) p < 0.05 0 % 1.14 to 1.71 0.678 None Critically low 5-HTTLPR Clarke et al. (2010) 2453/5600 39 Case-control OR 1.08(1.03,1.12) p < 0.05 44 % 1.03 to 1.27 0.39 None Critically low Not significant (NS) 5-HTTLpR S-Allele in the dominant model Cheng et al. (2014) 466/970 5 Cohort,
case-control,
cross-sectionalOR 1.98(0.92,4.26) p > 0.05 60 % 1.21 to 2.74 0.059 Large heterogeneity,
small study effectCritically low BDNF gene polymorphism-G vs A Bao et al. (2018) 2202/4638 7 Case-control OR 1.05(0.64,1.73) p > 0.05 74 % 0.55 to 1.54 0.196 Large heterogeneity Critically low BDNF gene polymorphism-GG vs GA Bao et al. (2018) 608/1594 7 Case-control OR 0.97(0.63,1.51) p > 0.05 0 % 0.53 to 1.40 0.674 None Critically low BDNF gene polymorphism-GG vs GA + AA Bao et al. (2018) 608/2319 7 Case-control OR 0.82(0.56,1.21) p > 0.05 7 % 0.43 to 1.20 0.506 None Critically low BDNF gene polymorphism-GG vs AA Bao et al. (2018) 608/1333 7 Case-control OR 0.63(0.37,1.07) p > 0.05 25 % 0.09 to 1.16 0.735 None Critically low
3.6Risk factors for bipolar disorder
Our detailed review of environmental risk factors for BD (Table 2) covers research findings from four eligible papers. Each of these four studies assess six potential environmental risk factors for BD, thus contributing to a comprehensive analysis in this field. The median case count across these studies was 3,693, with a range of 586 to 91,593, and a total of 124,365 cases. The overall number of participants involved in these studies was 3,687,969, with a median of 47,464, and a total sample size of 11,063,183,539. Regarding the statistical significance of the findings (Table 7, Fig. 6, e Appendix 12), the p-value for the random effects model was 0.001 for three (50 %) of the factors, and 0.05 for the remaining three (50 %) factors. With respect to heterogeneity, four of the six factors (66.67 %) had an I2 of more than 50 %, indicating moderate heterogeneity in these studies. Each of the studies had an Egger's value greater than 0.1, indicating an absence of publication bias. Our analysis did not reveal any weak study effects or significant biases.Study Number of case/total population Number of studies Study design Effect size Estimate
(95 %CI)Random effects
p valueI2 95 % prediction
intervalEgger's
P valueLarge heterogeneity,
small study effect,
excess significance
biasAMSTAR2 quality/
AMSTAR2 quality
when protocol
assessment was
ruled outSuggestive (class Ⅲ) Marijuana Gibbs et al. (2015) 5520/9373 2 Cohort RR 2.97(1.80,4.90) p < 0.001 0 % 2.46 to 3.47 0.326 None Critically low Rheumatoid arthritis Charoenngam et al. (2019) 91593/407115 6 Cohort,
case-controlRR 2.06(1.34,3.17) p < 0.001 86 % 1.62 to 2.49 0.09 Large heterogeneity;
small study effectCritically low Obesity Zhao et al. (2016) 12259/627749 9 Cross-sectional OR 1.77(1.40,2.23) p < 0.001 82 % 1.53 to 2.00 0.019 Large heterogeneity;
small study effectCritically low Weak (class Ⅳ) Perinatal asphyxia Shintani et al. (2023) 639/1281 5 Cross-sectional,
case-control,
cohortOR 1.46(1.02,2.11) p < 0.05 0 % 1.10 to 1.81 0.03 Small study effect High Obstetric complications Shintani et al. (2023) 586/1106 6 Case-control OR 1.41(1.18,1.69) p < 0.001 0 % 1.23 to 1.58 0.51 None High Birth weight less than 2500 g Shintani et al. (2023) 1865/85555 5 Case-control OR 1.28(1.04,1.56) p < 0.05 0 % 1.07 to 1.48 0.18 None High
3.7Biomarkers of bipolar disorder
In an extensive and detailed review (Table 2, Table 8, Fig. 7, e Appendix 13), seven eligible papers were analyzed to identify potential biomarkers of BD. These papers collectively assess 15 potential biomarkers, with a total case count of 2680. The median case count per study was 99, with a range of 52–548. The total number of participants across these studies was 13,458, with a median per study of 545, and a range of 104 to 4598.Study Number of case/total population Number of studies Study design Effect size Estimate
(95 %CI)Random effects
p valueI2 95 % prediction
intervalEgger's
P valueLarge heterogeneity,
small study effect,
excess significance
biasAMSTAR2 quality/
AMSTAR2 quality
when protocol
assessment was
ruled outWeak (class Ⅳ) DNR/RNR damage Brown et al. (2014) 117/230 4 Case-control Hedges' g 3.13(1.42,4.84) p = 0.0003 94 % 1.42 to 4.83 0.12 Large heterogeneity Llow Homocysteine Salagre et al. (2017) 72/495 9 Cross-sectional Hedges' g 0.98(0.80,1.17) p < 0.001 0 % 0.80 to 1.15 0.42 None Moderate Nitric oxide Brown et al. (2014) 103/356 6 Case-control Hedges' g 0.93(0.05,1.82) P=0.037 93 % 0.05 to 1.80 0.69 Large heterogeneity Low BDNF Fernandes et al. (2011) 548/1113 13 Cross-sectional SMD 0.81(0.52,0.11) p < 0.001 20 % 0.52 to 1.09 0.71 None Low Serum S100B Maria et al. (2016) 52/104 2 Case-control SMD 0.80(0.39,1.20) p < 0.001 0 % 0.39 to 1.56 0.11 None Low Elevated cerebrospinal fluid protein Sonja et al. (2018) 53/1457 2 Case-control SMD 0.77(0.36,1.18) p=0.0002 0 % 0.36 to 1.17 0.76 None Low F2-isoprostanes Maria et al. (2016) 106/729 8 Case-control,
cohortHedges' g 0.48(0.19,0.77) p = 0.001 73 % 0.19 to 0.76 0.13 Large heterogeneity Low Increased cerebrospinal fluid serum albumin ratio Sonja et al. (2018) 302/4598 4 Case-control SMD 0.43(0.25,0.61) p < 0.001 0 % 0.25 to 0.60 0.38 None Low Morning cortisol levels Girshkin et al. (2014) 67/776 18 Case-control Hedges' g 0.21(0.05,0.36) p = 0.008 0 % 0.05 to 0.36 0.47 None Critically low Folic acid level Hsieh et al. (2019) 481/1241 6 Case-control Hedges' g −0.21(-0.39,-0.03) p = 0.021 42 % −2.01 to 1.58 0.226 None Critically low Not significant (NS) Nitrotyrosine Brown et al. (2014) 90/190 3 Case-control Hedges' g 1.17(-0.16,2.50) p = 0.082 93 % −0.15 to 2.49 0.22 Large heterogeneity Critically low Protein carbony Brown et al. (2014) 99/454 5 Case-control Hedges' g 0.62(-0.40,1.64) p = 0.232 96 % −0.39 to 1.63 0.6 Large heterogeneity Low Superoxide dismutase Brown et al. (2014) 440/816 12 Case-control Hedges' g 0.12(-0.82,1.07) p = 0.802 97 % −0.81 to 1.05 0.93 Large heterogeneity Low Glutathione peroxidase Brown et al. (2014) 72/545 8 Case-control Hedges' g −0.05(-0.47,0.36) p = 0.815 79 % −0.46 to 0.36 0.46 Large heterogeneity Critically low Catalase Brown et al. (2014) 78/354 5 Case-control Hedges' g −1.58(-3.46,0.30) p = 0.098 98 % −3.45 to 0.29 0.76 Large heterogeneity Critically low
3.8Neurocognitive markers of bipolar disorder
In a focused review (Table 2, Table 9, Fig. 8, e Appendix 14), four papers were rigorously selected to investigate the neurocognitive markers associated with BD. These papers collectively explore five potential neurocognitive markers, with a substantial total of 20,745 case counts. The median case count per study was 2,197, with a range of 251–15,752 cases. The total number of participants involved in these studies was 64,085, with a median per study of 3,055, and a range of 396 to 50,563.Study Number of case/total population Number of studies Study design Effect size Estimate (95 %CI) Random effects
p valueI2 95 % prediction interval Egger's
P valueLarge heterogeneity,
small study effect,
excess significance
biasAMSTAR2 quality/
AMSTAR2 quality
when protocol
assessment was
ruled outWeak (class Ⅳ) ANK3 Single-nucleotide polymorphism Roby et al. (2017) 15752/50563 11 Case-control OR 1.18(1.06,1.31) SMD = 0.091(0.033,0.150) p < 0.05 68 % 1.07 to 1.28 0.32 Large heterogeneity Critically low S100 Calcium-binding protein B Bartoli et al. (2020) 348/581 8 Case-control SMD 0.81(0.36,1.26) p < 0.000001 81 % 3.49 to 3.84 0.001 Large heterogeneity,
small study effectLow Not significant (NS) GSK3B-50C/T Chen et al. (2014) 1251/3055 6 Case-control OR 1.06(0.99,1.14) SMD = 0.032(-0.007,0.071) p > 0.05 0 % 1.05 to 1.07 0.605 None Low Val66Met variant of BDNF Kanazawa et al. (2007) 3143/9490 11 Case-control OR SMD = −0.028(-0.069,0.012) p > 0.05 0 % 0.87 to 1.02 0.253 None Low Neuron-specific enolase Bartoli et al. (2020) 251/396 5 Case-control SMD −0.32(-1.02,0.38) p > 0.05 89 % 3.37 to 3.74 0.017 Large heterogeneity,
small study effectLow
4Discussion
4.1Summary of the critical findings and strengths of this study
This umbrella review is the first to methodically evaluate a wide range of factors associated with mood disorders, including protective factors, environmental risk factors, biomarkers, and neurocognitive markers. This study not only consolidates findings from past research but also introduces a robust quantitative analysis evaluating evidence strength. This review synthesizes data from 103 meta-analyses with a total of over 20 million participants from more than 2500 primary studies. Employing a consistent approach, this umbrella review reaches definitive conclusions on the strength of evidence for 134 factors and markers associated with depression and 26 associated with BD. The evidence supporting the associations between four environmental risk factors and MDD is of high credibility (i.e., class I evidence): history of mental illness, perinatal domestic violence, job strain, and being female. Furthermore, the associations between two environmental protective factors and MDD also showed evidence of high credibility (i.e., class I evidence): Mediterranean diet score (MDS) and green tea consumption.
4.2Risk factors for major depressive disorder
Gender (male or female) significantly influences the risk of major depressive disorder, with a higher prevalence of depression among women than men (Lopez Molina et al., 2014). This disparity, which has been attributed to physiological and hormonal factors, is evident in the various life stages (Halbreich and Kahn, 2001; Jung et al., 2015). There are higher depression rates among adolescent girls than adolescent boys, and women experience specific mood-related disorders, such as premenstrual dysphoric disorder, postpartum depression, and perimenopausal depression, that are exclusive to the female gender (Angst et al., 2002; Soares and Zitek, 2008). Hormonal factors (Soares and Zitek, 2008), including ovarian gonadotropins and neurotransmitters modulated by ovarian hormones, play a crucial role in the neurobiology of depression in women (Duclot and Kabbaj, 2015; O'Hara and McCabe, 2013). The Effort–Reward Imbalance model demonstrates that job strain, which is characterized by low reward and high effort, elevates the risk of depression (Borrow and Cameron, 2014). In addition, disruptions in circadian rhythm and night shift work contribute to an elevated likelihood of depression. These findings underscore the multifaceted nature of the risk factors for depression, especially in relation to gender-specific physiological and environmental influences (Devries et al., 2013).
4.3Protective factors for major depressive disorder
Adherence to the Mediterranean diet (MD) is increasingly recognized as a preventive measure against depression. The World Health Organization has documented a rise in the prevalence of noncommunicable diseases such as diabetes, obesity, and cardiovascular disease (Valkanova and Ebmeier, 2013; Nouwen et al., 2010; Jokela et al., 2014), which are now responsible for 70 % of deaths globally and are considered risk factors for depression and other mental health conditions (Iriti et al., 2020; Valkanova and Ebmeier, 2013). MD is characterized by an emphasis on moderate consumption of fish, dairy, and poultry, coupled with meeting daily intake recommendations for vegetables, fruits, and cereals, as well as water, tea, red wine, and coffee (Saghafian et al., 2023; Grosso et al., 2016; Li et al., 2016). MD and related dietary practices have been demonstrated to be protective against depression, with the consumption of tea and fruits backed by the strongest evidence (Saghafian et al., 2023; Grosso et al., 2016; Li et al., 2016; Liu et al., 2016; Luo et al., 2023). Many studies emphasize the vital role of diet in the onset of depression, which has fueled interest in dietary approaches for the prevention and treatment of depression (Sarris et al., 2015; Molendijk et al., 2018). The anti-inflammatory and antioxidant properties of MD are believed to enhance the immune system, improve reproductive health, prevent mental health issues, reduce the risk of depression, and decrease the likelihood of neurodegenerative diseases (Gantenbein and Kanaka-Gantenbein, 2021).
4.4Biomarkers of major depressive disorder
Brain-derived neurotrophic factor (BDNF), identified as a potential biomarker for depression, is supported by suggestive evidence of its role in this context. Given the current lack of clarity regarding the precise pathophysiological mechanisms underlying depression, evaluating its biomarkers could play a critical role in facilitating increasingly accurate assessments of the depressive states of patients in the future. Numerous studies, including those involving autopsies, animal models, and clinical settings, have reported lower BDNF levels in individuals with depression (Emon et al., 2020; Lorenzetti et al., 2020; Murínová et al., 2017; Nunes et al., 2018; Sheldrick et al., 2017). A comprehensive meta-analysis found significantly lower BDNF levels in depressed individuals compared to their non-depressed counterparts. However, it is important to note that although BDNF is a potential biomarker for depression, its reliability in this capacity has not yet been firmly established (Tiwari et al., 2022).
4.5Neurocognitive markers of major depressive disorder
Methylenetetrahydrofolate reductase (MTHFR), an enzyme integral to folate metabolism in the body, has been identified as a significant neurocognitive marker for depression (Jiang et al., 2016). The earliest association of MTHFR C677T polymorphism with depression was reported in a Japanese study (Arinami et al., 1997; Lin and Tseng, 2015). Subsequently, a meta-analysis highlighted that all variants of MTHFR C677T are linked to an elevated risk of depression, especially among the Chinese population. Notably, low folate concentrations, which are more prevalent in northern populations, are associated with a high incidence of depression. These findings underscore the importance of decreased MTHFR activity and folate deficiency as potential contributing factors to depression (Jiang et al., 2016).
4.6Risk factors for bipolar disorder
In a clinical study of BD patients, a notable causal relationship was observed between manic symptoms and marijuana use (Henquet et al., 2006), with a significant correlation between the duration of manic episodes and active marijuana use (Tijssen et al., 2010). In addition, a meta-analysis reports a pooled effect size, indicating that cannabis use exacerbates BD symptoms (Gibbs et al., 2015). Tetrahydrocannabinol, a key component of cannabis, enhances dopamine neural pathways, increases dopamine activity, and inhibits dopamine uptake via the cannabinoid receptor. This dopamine overactivity leads to dysregulation of the dopamine system, a known pathological mechanism of mania (Murray et al., 2004; D'Souza et al., 2005). However, it is crucial to note that this risk factor has not been proven definitively due to the limited number of similar clinical studies and the dearth of studies that meet the inclusion criteria.
4.7Biomarkers of bipolar disorder
Rheumatoid arthritis (RA), an autoimmune condition that primarily affects the joints, also has implications for other systems in the body (Xue et al., 2020). Emerging research indicate a potential link between RA and an elevated risk of BD. The findings of several studies point to this association, and a meta-analysis further supports this finding (Cremaschi et al., 2017; Forty et al., 2014; Marrie et al., 2018; Wang et al., 2018). The meta-analysis showed that individuals with RA had a higher prevalence of BD compared to individuals who did not have RA, and RA is associated with an elevated risk of developing BD (Charoenngam et al., 2019). Furthermore, depression is often found to be comorbid with RA, with depression being a key symptom of RA. However, it is important to note that although arthritis is a well-supported risk factor for depression, there is currently only suggestive evidence of arthritis as a risk factor for BD. This potentially may be due to the greater impact arthritis has on depression than on BD, or the lack of extensive research in this area.
4.8Neurocognitive markers of bipolar disorder
There are several other biomarkers and neurocognitive markers with an evidence base that remains weak or insignificant. For instance, ANK3 single nucleotide polymorphism and S100 calcium-binding protein B. Although the evidence synthesized in this umbrella review indicates that the ANK3 gene single nucleotide polymorphism and S100 calcium-binding protein B have been proposed as potential neurocognitive markers for BD, the supporting basis still has significant limitations (Charoenngam et al., 2019). This limitation stems largely from a dearth of studies and case reports elucidating the impact of these markers on depression. Furthermore, a significant number of meta-analyses in this domain have not adequately addressed the risk of bias or utilized protocol registries, further undermining the reliability of their findings.
4.9Limitations
Our study has several limitations. First, we included only the English-language literature in our umbrella review, which may have introduced some bias into our results, thus limiting the comprehensiveness of our study findings. This issue should be addressed in future research by endeavoring to incorporate studies published in other languages that are relevant to the research topic. Second, the observational design of the analyzed studies allows for stable associations but does not rule out potential confounding factors, making it challenging to establish causality between factors such as gender, depression, work stress, and perinatal domestic violence (Howard et al., 2013; Madsen et al., 2017; Tang et al., 2014). Third, in our analysis, we could not differentiate between various types of mood disorders, limiting the depth of our disorder-specific insights. Fourth, the observed associations with neurocognitive markers are weak, indicating the need for more research in this area. Fifth, the identified factors may overlap, and our dependence on published meta-analyses may have led to certain environmental factors or biomarkers being overlooked or underrepresented. Sixth, the absence of systematic reviews on environmental protective factors for BD points to a gap in the current literature. Seventh, the gender mentioned in our research results refers to the binary gender identity (male or female) reported by the participants themselves. This limitation is mainly based on the descriptions and data provided in the original literature we included. To faithfully represent the original research, we have retained the term “gender” and ensured the traceability of the results. Future research needs to adopt more inclusive methods for measuring gender. Finally, although the evidence for some factors and biomarkers is strong, they do not confirm a causal relationship with the disorders, indicating a need for further advanced validation research. Notwithstanding these limitations, our findings reinforce existing clinical guidelines that emphasize the comprehensive management of mood disorders and highlight the significance of environmental influences, dietary patterns, and lifestyle choices. In particular, our findings support gender-specific approaches to treating depression and caution against cannabis use in BD management. These insights encourage clinicians to integrate the various factors evaluated in this review into personalized treatment plans, underscoring a holistic approach to addressing the complexities of mood disorders in clinical practice (Choi and Kang, 2022; Xue et al., 2020).
5Conclusion
This umbrella review methodically amalgamates a vast range of meta-analyses and provides a thorough assessment of environmental risk factors, protective elements, biomarkers, and neurocognitive markers relevant to depression and BD. Based on high-quality evidence, we have identified priority intervention targets for MDD, such as the need to promote the construction of workplace psychological protection systems in response to work stress and the call to strengthen the support network for pregnant women in response to perinatal domestic violence. At the same time, the Mediterranean diet and green tea intake, as strong protective factors, support their inclusion in public health nutrition strategies. For BD, high-risk factors such as rheumatoid arthritis suggest the need for regular mood monitoring in patients with autoimmune diseases. At the level of precision prevention, for example, the MTHFR C677T polymorphism as a neurocognitive marker for MDD lays the foundation for the construction of a gene-environment interaction risk assessment model, and future individualized intervention can be achieved through targeted folic acid supplementation. Importantly, BDNF as a common biomarker for both diseases reveals the core role of neurotrophic pathways in mood disorders, advocating for the development of cross-disease intervention programs to significantly enhance prevention efficacy. In summary, this study provides clinical significance from evidence to practice for the early detection and targeted intervention of the diseases.
Data sharing
All data included in this umbrella review were extracted from publicly available meta-analysis.
Declaration of competing interest
We declare no competing interests.
Appendix ASupplementary data
The following is the Supplementary data to this article:
Data availability
Data will be made available on request.
Acknowledgements
This work was supported by the Anhui Province University Scientific Research Projects (2023AH040086), and Key Laboratory of Philosophy and Social Science of Anhui Province on Adolescent Mental Health and Crisis Intelligence Intervention (SYS2023B08), and 10.13039/501100001809National Natural Science Foundation of China (82090034).